##### Question You are interviewing for a Data Scientist role on a marketing/growth team. Sales has limited outreach capacity, so the business wants a **lead-scoring** system that ranks or scores incoming leads (a user or account arriving through ads, email, organic, etc.) so Sales/Marketing can prioritize who to contact. Assume you have a historical dataset of leads with a `lead_id`, a `created_at` timestamp, features available at scoring time (acquisition channel/campaign/geo/device, firmographics such as company size and industry, behavioral signals such as pages viewed, pricing-page hits, demo requests, email engagement), and one or more outcome labels (e.g. `converted` within a defined window, and optionally `time_to_convert_days`). Design an end-to-end approach. Be explicit about assumptions (conversion window, label definition, scoring cadence) and call out key pitfalls and edge cases. 1. Define the prediction **target (label)** and the **prediction time** (when the score is computed). Address how you handle leads that are too recent to have observed the outcome window. 2. Propose **feature sets and data sources**, and explain how you would handle feature availability and **leakage**. 3. Propose both a **statistical (baseline) model and a more advanced machine-learning model**, and explain the interpretability/performance tradeoffs. 4. The stakeholder may either only care about predictive performance, or require understanding **which features are important and why**. Explain what you would deliver in each scenario. 5. Explain what **multicollinearity** is, why it matters (or doesn't) for different model families, how you would detect it, and how you would mitigate it. 6. Define how you would **evaluate** the model: a primary metric (and why), diagnostic metrics/plots, and guardrails (fairness, stability, operational constraints). Tie metrics to business constraints such as top-K capacity, lift, calibration, and revenue. 7. Describe how you would pick an **operating threshold / routing policy** to turn scores into actions. 8. Discuss key **risks** — class imbalance, selection bias (sales touches are not random), fairness, and drift — and how you would monitor and iterate. 9. Describe how you would **deploy and monitor** the score in production and how you would update it over time.

Design an end-to-end lead-scoring model for a marketing/growth data science team: define the label and scoring time, engineer leakage-safe features, compare a logistic-regression baseline with gradient-boosted trees, and evaluate with capacity-aware ranking metrics. The question also probes multicollinearity, model interpretability (SHAP/reason codes), routing-threshold design, and operational risks like selection bias, class imbalance, fairness, and drift.

How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

What difficulty level is this interview question?

This is a easy difficulty Machine Learning question, commonly asked during Technical Screen rounds at IBM.

What role is this question designed for?

This question is commonly asked for Data Scientist candidates at IBM during technical interviews.

Design a lead-scoring model | IBM Interview Question

Question

You are interviewing for a Data Scientist role on a marketing/growth team. Sales has limited outreach capacity, so the business wants a lead-scoring system that ranks or scores incoming leads (a user or account arriving through ads, email, organic, etc.) so Sales/Marketing can prioritize who to contact.

Assume you have a historical dataset of leads with a lead_id, a created_at timestamp, features available at scoring time (acquisition channel/campaign/geo/device, firmographics such as company size and industry, behavioral signals such as pages viewed, pricing-page hits, demo requests, email engagement), and one or more outcome labels (e.g. converted within a defined window, and optionally time_to_convert_days).

Design an end-to-end approach. Be explicit about assumptions (conversion window, label definition, scoring cadence) and call out key pitfalls and edge cases.

Define the prediction target (label) and the prediction time (when the score is computed). Address how you handle leads that are too recent to have observed the outcome window.
Propose feature sets and data sources , and explain how you would handle feature availability and leakage .
Propose both a statistical (baseline) model and a more advanced machine-learning model , and explain the interpretability/performance tradeoffs.
The stakeholder may either only care about predictive performance, or require understanding which features are important and why . Explain what you would deliver in each scenario.
Explain what multicollinearity is, why it matters (or doesn't) for different model families, how you would detect it, and how you would mitigate it.
Define how you would evaluate the model: a primary metric (and why), diagnostic metrics/plots, and guardrails (fairness, stability, operational constraints). Tie metrics to business constraints such as top-K capacity, lift, calibration, and revenue.
Describe how you would pick an operating threshold / routing policy to turn scores into actions.
Discuss key risks — class imbalance, selection bias (sales touches are not random), fairness, and drift — and how you would monitor and iterate.
Describe how you would deploy and monitor the score in production and how you would update it over time.

Question

Design an end-to-end approach. Be explicit about assumptions (conversion window, label definition, scoring cadence) and call out key pitfalls and edge cases.

Define the prediction target (label) and the prediction time (when the score is computed). Address how you handle leads that are too recent to have observed the outcome window.
Propose feature sets and data sources , and explain how you would handle feature availability and leakage .
Propose both a statistical (baseline) model and a more advanced machine-learning model , and explain the interpretability/performance tradeoffs.
The stakeholder may either only care about predictive performance, or require understanding which features are important and why . Explain what you would deliver in each scenario.
Explain what multicollinearity is, why it matters (or doesn't) for different model families, how you would detect it, and how you would mitigate it.
Define how you would evaluate the model: a primary metric (and why), diagnostic metrics/plots, and guardrails (fairness, stability, operational constraints). Tie metrics to business constraints such as top-K capacity, lift, calibration, and revenue.
Describe how you would pick an operating threshold / routing policy to turn scores into actions.
Discuss key risks — class imbalance, selection bias (sales touches are not random), fairness, and drift — and how you would monitor and iterate.
Describe how you would deploy and monitor the score in production and how you would update it over time.

Design a lead-scoring model

Quick Overview

Question

Solution

Submit Your Answer to Earn 20XP

Design a lead-scoring model

Quick Overview

Question

Solution

Submit Your Answer to Earn 20XP